WhiteMagic — governed local memory for agent continuity

WhiteMagic meta-tool — memory and continuity kernel over the curated tool surface (70 tools): persistent memory, session continuity and recall, governance/audit, and local tool execution. Mode: READ-ONLY — writes (session.*, memory.create/update) are refused; recall only. Scope: store /srv/whitemagic/hosted-store/lmdb. Invoke with thought=<natural language> (auto-routed), route=<exact tool id> (e.

Search memories: hybrid BM25 + vector fusion with deterministic episodic and full-text fallbacks; each result discloses its recall_mode. The preferred recall tool.

Retrieve a specific memory by its ID or coordinate address.

List memories with filters (galaxy, tags, limit, offset) for browsing and discovery.

Compatibility alias of memory.search — same implementation, same arguments, same results. Kept for clients pinned to this name; prefer memory.search.

Recall where the previous session left off: recent turns, checkpoints, and open work.

List or verify a memory's content revision chain (tamper evidence): every content update is chained and head-checked.

Verify a receipt id or a bundle offline; fail-closed without an anchor (verdict PROVISIONAL).

Count memories matching optional filters without returning content.

Memory index statistics: totals, galaxy spread, and index health.

List the tag vocabulary with usage counts.

Aggregate a numeric or categorical field over matched memories.

Associative neighbours of a memory (spreading activation over the association graph).

Read several memories by id in one call.

Structured query over memories with filters and ordering.

Filter memories by metadata (tags, importance, recency).

Memories near a coordinate address (5D spatial neighbourhood).

Vector-similarity search over memory embeddings.

List sessions with titles and summaries.

Recall a session's turns without continuity packaging.

Replay a session's recorded turns in order.

Knowledge-surface status: galaxies, counts, and index health.

Explain why a memory or route was selected (provenance readout).

Execute multiple memory search queries concurrently in a single turn. Unpacks up to 10 queries, searches the curated corpus, and returns consolidated results.

Request a small uncollateralized compute advance. Phase 1 (2026-10-15): bound to your authenticated evaluation key (daily per-IP and global caps); new key identities start at a $0.10 ceiling that grows $0.25 per verified settled payment. Returns a recall-scope session pass; hosted write requires a paid lease.

Settle outstanding compute balance to upgrade Bayesian reputation score and dynamically expand credit ceiling. Supports payment reference or gasless EIP-3009 transfer authorization.

Inspect public transparency metrics: total loans disbursed, default rates, zero predatory liquidation policy, and sovereign free invariants.

Zero-latency provably fair mutex lock and resource contention arbitration between 2 or more competing autonomous agents. Resolves locks in microseconds without LLM debate token waste.

Cryptographically secure commit-reveal dice engine for autonomous agent symmetry breaking, randomized exploration, or provably fair decision making.